Modeling networking protocols to test intrusion detection systems

Techniques for detecting previously unseen network intrusion attempts often depend on finding anomalous behavior in network traffic streams. Thus, intrusion detection systems generally require tuning to be effective in each new environment. It follows that researchers need to produce traffic backgrounds for laboratory testing that accurately reflect the characteristics of organizations of interest. To these they need to be able to add attacks under controlled conditions so that actual performance of new IDS techniques can be evaluated. The authors have been working towards such a realistic and flexible testing environment. In previous work we have used application protocol traffic generation techniques and confirmed some of the classical statistical distributions. The distributions of other protocols have been updated based on the analysis of current traffic. In This work hybrid and heavy-tailed modeling techniques are used to build a detailed model of FTP including session arrivals, bytes transferred, and idle times. The same techniques are being used for other protocols including HTTP.